Nimble MCP Server — Enterprise Web Intelligence With the Best Google Maps Tools

Part of our Web Search & Data Extraction MCP category.

At a glance: Closed-source MCP server/proxy backend (hosted only) with a separate MIT-licensed open-source agent-skills SDK, 18 tools (expanded from 7), Streamable HTTP transport, free tier (5,000 requests/month), pay-as-you-go from $1/1,000 requests, managed plans from $2,500/month. $47M Series B (February 2026). Databricks Marketplace one-click install.

Nimble (formerly Nimbleway) is an enterprise web data platform that raised $47 million in Series B funding in February 2026 to build AI agent infrastructure. Databricks Ventures participated in that round, and Nimble is listed on the Databricks Marketplace as a one-click Unity Catalog connection — making Nimble simultaneously an investor-backed and distribution-channel partner of Databricks. Their MCP server connects to the same proxy and scraping backend used by enterprise customers, wrapping it in a standard MCP interface so agents can pull structured web data in real time.

Since the original review, Nimble has significantly expanded its MCP surface: 7 tools have grown to 18, the transport has migrated from SSE to Streamable HTTP, and a new “Web Search Skills” layer launched in April 2026 with 10 pre-built vertical workflows.

The standout feature remains: Nimble ships three dedicated Google Maps intelligence tools — business discovery, venue details (including real-time crowd levels), and structured review collection with sentiment analysis, in one native suite. It’s not the only MCP path to Google Maps data anymore — Google itself shipped an official Maps Grounding Lite MCP server (routing, weather, points-of-interest) in April 2026, and community/Apify-built Google Maps scraper MCP servers exist — but among the MCP servers this site has reviewed, Nimble is still the only one offering a native, purpose-built suite with sentiment analysis and real-time crowd-level data.

The key question remains: at $2,500/month minimum for managed plans, is this justified when Bright Data offers 60+ tools with a free tier, and Firecrawl handles general scraping at a fraction of the cost?

What It Does

Nimble’s MCP server now exposes 18 tools across five categories, expanded from the original 7 in three categories.

Web Search & Extraction

deep_web_search doesn’t just return search results — it follows links and extracts the actual page content. A meaningful difference from SERP-only tools. The nimble_extract tool picked up screenshot capture (base64 PNG) in February 2026.

Note (March 2026 breaking change): HTML output is no longer returned from Extract by default — it now requires an explicit formats parameter opt-in. HTTP response headers are also excluded by default. Existing pipelines relying on HTML or headers need updating.

Map & Crawl (New in 2026)

These two tools fill a significant gap from the original review. nimble_map is the equivalent of Firecrawl’s map tool — useful for understanding site structure before targeted scraping. nimble_crawl handles progressive multi-page extraction with depth controls, comparable to Firecrawl’s crawl capability.

Custom Agent Builder (New in 2026)

This category is Nimble’s most differentiated new addition. Agents are reusable extractors — you define a schema once and Nimble applies it to any URL, returning structured JSON output matching your spec. The Claude Code plugin exposes this as an “Agent Builder” skill that can generate batch scripts and preview schemas interactively. Batch processing up to 1,000 agent requests in a single API call is supported via the underlying API.

E-commerce & Targeted Scraping

The targeted_engines tool acts as a discovery layer — your agent can ask what platforms are supported before attempting to scrape. The targeted_retrieval tool returns structured JSON with product data, not raw HTML.

Google Maps Intelligence

This remains Nimble’s strongest differentiator among the servers this site has reviewed. For agents doing local business research, competitive analysis, or location intelligence, these three tools fill a real gap. It is not a total ecosystem monopoly, though — see the note above about Google’s own Maps Grounding Lite MCP server and Apify-based Google Maps scrapers, neither of which matches Nimble’s combination of crowd-level data and review sentiment analysis in one native suite.

Web Search Skills (April 22, 2026)

Alongside the MCP tools, Nimble launched 10 vertical “Web Search Skills” as a higher-level workflow layer. These aren’t raw MCP tools — they’re pre-built research workflows built on the same infrastructure, designed to run interactively or in autonomous scheduled mode:

• competitor-intel, company-deep-dive, market-finder, competitor-positioning, seo-intel
• meeting-prep, local-places
• healthcare-providers-extract, healthcare-providers-enrich, healthcare-providers-verify
• talent-sourcer

The Skills layer is available via Claude Code (claude.com/plugins/nimble) and as a standalone, MIT-licensed open-source SDK (Nimbleway/agent-skills, v1.6.0 as of August 2026 — actively released, so check the releases page for the current tag). Skills accept natural-language inputs and return structured research reports — closer to what LangChain’s agent frameworks provide than to raw MCP tool calls. For enterprise teams that don’t want to build their own workflows, this is a meaningful addition.

Benchmark Performance

In testing by AIMultiple across multiple MCP servers (disclosure: Bright Data, Apify, and Oxylabs are disclosed sponsors of the AIMultiple benchmark page; Nimble is not a listed sponsor):

Web Search & Extraction (single-agent):

Nimble ranked second in accuracy (93%) behind only Bright Data’s perfect score. Response time was moderate at 16 seconds — faster than Bright Data but slower than Firecrawl.

Stress Test (250 Concurrent Agents):

The concurrent load weakness has not been publicly retested or resolved. Under 250 simultaneous agents, Nimble’s success rate dropped to 51.2% (down from 93%), and successful tasks averaged 182.3 seconds — nearly 4x slower than Oxylabs. The new batch endpoint (up to 1,000 agent requests per call) operates at the API layer but doesn’t address the per-request latency and reliability degradation at scale. If high-volume parallel agent workflows are your use case, this remains a meaningful concern.

Setup

Nimble migrated from SSE to Streamable HTTP transport. The old SSE endpoint (/sse) is no longer the official target — connect via the MCP endpoint instead:

{
  "mcpServers": {
    "nimble": {
      "url": "https://mcp.nimbleway.com/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_NIMBLE_API_KEY"
      }
    }
  }
}

Via Claude Code CLI:

claude mcp add --transport http nimble-mcp-server https://mcp.nimbleway.com/mcp

Works with Claude Desktop, Cursor, qodo, Microsoft CoPilot, OpenAI Playground, Google ADK, Databricks, and more. Framework integrations for LangChain, LlamaIndex, Agno, and Microsoft AutoGen are documented.

No self-hosting option. All requests route through Nimble’s infrastructure. This simplifies setup but means you’re dependent on their uptime, pricing, and data handling.

Pricing

Updated 2026-08-18 — the pay-as-you-go rates below have changed materially since the original review; see Nimble’s live pricing page.

Pay-as-You-Go

Managed Plans

All managed plans include MCP integration, custom agent ETL, and 15% discount on annual billing. The managed-plan tiers and prices are unchanged since the original review, but the pay-as-you-go rate card above has been restructured and, for search, roughly 5x more expensive per request than at launch.

Free tier: 5,000 requests per month, no credit card required — as of this audit (August 2026) this is described as an ongoing monthly allowance, not the one-time trial the original review reported. That’s the same headline size as Bright Data’s free tier below, so the free-tier gap between the two has closed.

The pricing gap vs. alternatives is now mainly about the managed tier. Bright Data offers a free tier with 5,000 requests/month on an ongoing basis — matching Nimble’s. Firecrawl has paid plans starting at $19/month. Nimble’s managed tier starts 130x higher than Firecrawl’s entry point.

Security & Compliance

Nimble holds AICPA SOC 2 certification and claims GDPR and CCPA alignment. Infrastructure runs on AWS and GCP with role-based access controls. The platform now explicitly markets zero data retention and PII masking as security features — new positioning language since the prior review, useful for enterprise procurement conversations.

No new certifications (ISO 27001, FedRAMP) have been announced.

How It Compares

Limitations

• No open source, no self-hosting. You’re fully locked into Nimble’s infrastructure and pricing. If they raise prices or change terms, you have no alternative deployment option.
• Stress test performance still unresolved. 51.2% success at 182.3 seconds under concurrent load makes this unsuitable for high-volume parallel agent workflows, and no public fix or retest has been announced.
• Expensive entry point for the managed tier. $2,500/month minimum for managed plans — unchanged since February 2026 — even though the pay-as-you-go and free tiers are now more competitive than at launch.
• Closed-source trust model. No way to audit how your data is handled, what’s cached, or how requests are routed. The SOC 2 certification partially addresses this, but transparency-minded teams may prefer open-source alternatives.
• The core MCP server/proxy backend has no public GitHub repo. You can’t audit or self-host the piece that actually does the scraping and proxying. There is a separate, small, MIT-licensed agent-skills SDK repo (issues enabled, ~50 stars, ~16 forks as of August 2026) for the Web Search Skills layer, but that’s a thin client on top of the closed backend, not the MCP server itself — you still can’t self-host or fully audit request handling.
• Limited grassroots traction. Nimble MCP is absent from most independent “best MCP servers” roundup articles and has minimal Reddit or HackerNews discussion — Bright Data and Firecrawl dominate community mindshare.
• Breaking change: HTML no longer returned by default from Extract (March 2026) — existing pipelines using the API directly may need updates.

The Bottom Line

Nimble has shipped meaningful improvements since its March launch: the tool count jumped from 7 to 18 with the addition of Map, Crawl, and custom Agent Builder categories; the transport migrated to Streamable HTTP; and the Web Search Skills layer adds pre-built vertical research workflows. The Databricks Marketplace listing gives enterprise teams a frictionless installation path.

The Google Maps intelligence suite remains Nimble’s clearest differentiator among reviewed MCP servers — three dedicated tools for business discovery, venue details, and review extraction with sentiment analysis. It’s no longer the only MCP path to Google Maps data (Google’s own Maps Grounding Lite MCP server and community/Apify scrapers exist), but no other reviewed server matches this specific combination.

What hasn’t changed: the concurrent-load stress test weakness (51.2% success, 182.3s average) remains a concern for high-volume workloads, and the $2,500/month managed plan floor remains one of the highest entry points in the category. Community adoption continues to lag behind Bright Data and Firecrawl.

Who should use this: Teams building agents that need local business intelligence (Google Maps data), e-commerce price monitoring across major retailers, enterprises that require SOC 2 compliance documentation, or Databricks users who want a one-click install. The new Agent Builder tools and Web Search Skills layer also suit teams that want higher-level research workflows without building their own.

Who should look elsewhere: Individual developers, teams on a budget, anyone needing browser automation or social media scraping, or high-concurrency agent deployments. Bright Data offers more tools with better stress performance at lower entry cost. Firecrawl handles general web scraping — including crawl and map — with an open-source self-hosted option starting at $19/month.

Rating: 3.5/5 — The platform grew substantially (7 → 18 tools, Streamable HTTP, Web Search Skills, Databricks integration), and the Google Maps intelligence suite remains its best differentiator among reviewed MCP servers (though no longer the only MCP path to Google Maps data). Stress test performance and pricing floor remain limiting factors for broad recommendation.